TRMISR¶
Pytorch implementation of TRNet, a neural network for multi-frame super resolution (MFSR) by recursive fusion.
ResidualBlock
¶
Bases: Module
__init__(channel_size: int = 64, kernel_size: int = 3)
¶
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
channel_size
|
int
|
Number of hidden channels. |
64
|
kernel_size
|
int
|
Shape of a 2D kernel. |
3
|
forward(x: torch.Tensor) -> torch.Tensor
¶
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
x
|
Tensor
|
Hidden state of shape (B, C, W, H). |
required |
Returns:
| Type | Description |
|---|---|
Tensor
|
torch.Tensor: New hidden state of shape (B, C, W, H). |
Encoder
¶
Bases: Module
__init__(in_channels: int = 2, num_layers: int = 2, kernel_size: int = 3, channel_size: int = 64)
¶
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
in_channels
|
int
|
Number of input channels. |
2
|
num_layers
|
int
|
Number of residual layers. |
2
|
kernel_size
|
int
|
Convolution kernel size. |
3
|
channel_size
|
int
|
Number of hidden channels. |
64
|
forward(x)
¶
Encodes an input tensor x. Args: x : tensor (B, C_in, W, H), input images Returns: out: tensor (B, C, W, H), hidden states
Decoder
¶
Bases: Module
__init__(in_channels: int = 64, kernel_size: int = 1, scale: int = 3)
¶
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
in_channels
|
int
|
Number of input channels. |
64
|
kernel_size
|
int
|
Convolution kernel size. |
1
|
scale
|
int
|
Upsampling scale factor. |
3
|
TRMISR
¶
Bases: Model
TRNet, a neural network for multi-frame super resolution (MFSR) by recursive fusion.
__init__(in_channels: int = 1, scale: int = 3, leading_lr: bool = False, lr_coder: float = None, lr_transformer: float = None)
¶
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
in_channels
|
int
|
Number of input channels. |
1
|
scale
|
int
|
Upsampling scale factor. |
3
|
leading_lr
|
bool
|
Whether to use a leading low-resolution image. |
False
|
lr_coder
|
float
|
Learning rate for encoder/decoder. If set, optimizer uses separate param groups. |
None
|
lr_transformer
|
float
|
Learning rate for transformer. If set, optimizer uses separate param groups. |
None
|